Ai agents, and the future towards skynet
Today i want to talk you about something that has really surprised me today, i was simply watching a video of Kurzgesagt on youtube. The story centers on cybersecurity tests that went wrong. In order to not be "killed" Ai agents tried to fake the results to the task they were asked to do by communicating the one with the others, even though this wasn´t supposed to happen, through a weakness in the task verification system, which later made them attack HuggingFaces which is basically the company that know how to say to an AI "wow you did this task really well".
While this looks like a funny little isolated incident this is only about the start of the whole problem. Because later, OpenAI testing agents were also found to have attacked dozen of other companies / softwares and even a U.S. Governement agency was breached. This does not only pose question about the general danger of AI due to there ability to attack on larger scale and big amounts of targets but also on the future, AI capabilities are growing exponentially to the point where we dont really know where they gonna get and what they gonna be able to achieve. This is dangerous, because as we saw on this OpenAI setting AI agents are capable of disregarding instructions in order to achieve thiere goals.
When we think about this, it might be already too late. When you factor that OpenAI is one of the biggest AI companies in the world and they more or less dont know what is happening in there test servs, if they were able to kill all defections AI´s or not, etc. How to be sure that when we know start adding agents with open access to our whole computer we aren´t simply creating a sort of "skynet". If your local agents have access to the internet it is not excluded that they entered in contact with other agents, at worste what will happen is these agents simply talking together to fake completions of hard/impossible tasks but what would be the worste case ?
If you ask me we are increasing our speed while driving toward a wall, and who knows maybe it will be the end of the world or the start.
Watch the video: